Pith. sign in

REVIEW 2 major objections 3 minor 68 references

Language Models Generalize to Human-like Word Order Preferences

T0 review · 2 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that language models trained on a corpus with every multi-modifier noun phrase removed still develop a reliable preference for scope-homomorphic modifier order, matching human artificial-language-learning results.

desk verdict The core FiCT result is real and worth engaging; the 'never observed' claim is only as strong as an unvalidated parser filter. read the letter →

arxiv 2608.05028 v1 pith:Z2A7SXW4 submitted 2026-08-05 cs.CL

Amanda Popadich , Shane Steinert-Threlkeld This is my paper · ORCID
classification cs.CL
keywords languagemodelsmodifierorderingscope-homomorphismartificiallearningpovertyofthestimulusfilteredcorpustrainingpointwisemutualinformationinductivebias
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether a general-purpose language model can acquire a human-like bias for ordering modifiers inside a noun phrase when the training data contains no noun phrases with more than one modifier. The authors build a 100-million-token English corpus in which every multi-modifier noun phrase detected by a dependency parser is rewritten into single-modifier variants, train decoder-only transformers of three sizes on it, and then test on minimal pairs that contrast the homomorphic order with the swapped order. All three models reliably prefer the homomorphic order—'two fluffy dogs' over 'fluffy two dogs'—even though they never saw such phrases in training. The preference is strongest for demonstrative-numeral combinations and weakest for demonstrative-adjective combinations, and it is not explained by noun-modifier association strength measured by pointwise mutual information. If the result holds, it suggests that a widely documented human learning bias can emerge from impoverished input in a general learning system, without an innate, language-specific constraint.

What carries the argument

The central mechanism is filtered corpus training (FiCT): every sentence containing a noun phrase with multiple target modifiers—demonstratives, numerals, and adjective blocks—is decomposed into variants that keep only one modifier per noun, so the training corpus contains only single-modifier noun phrases. The target generalization, scope-homomorphism, is the hypothesis that the linear order of modifiers mirrors their compositional semantic scope, so that wider-scope modifiers come first. Evaluation uses minimal pairs of the form $M_1 M_2 N$ versus $M_2 M_1 N$, scored by the log-probability difference $HC_\Delta(M_1M_2N|c) = \log P(M_1M_2N|c) - \log P(M_2M_1N|c)$ and by the proportion of items for which this difference is positive ($HC\%$). Positive values indicate a preference for the order that matches semantic scope. The same measure is then compared against per-item differences in pointwise mutual information between the noun and each modifier, which is the distributional account the paper tests and rejects.

What would settle it

Audit the training corpus (or the filtering pipeline) against a gold-standard set of noun phrases with multiple modifiers and count how many multi-modifier noun phrases survive the perturbation; if the count is nonzero and large enough to provide signal, the claim that models never observed the target structure is false.

Watch

Extended reading notes

Core claim

Under filtered-corpus training conditions that remove all direct evidence of multi-modifier noun phrase order, OPT-style decoder-only language models with 52M, 110M, and 350M parameters develop a statistically reliable preference for scope-homomorphic modifier orders. Across model sizes, homomorphic consistency accuracy is 70–76% and mean log-probability differences are positive, with the largest effects for demonstrative-numeral orderings. The models' modifier-type profile differs from humans—demonstrative-adjective is the strongest preference in human experiments but the weakest here—so the result is not a simple copy of the human pattern. Item-level analysis with pointwise mutual information shows that noun-modifier association strength does not predict model preferences, ruling out one distributional account of the bias. The paper concludes that the preference can be recovered from impoverished input, though the exact representational or statistical source remains open.

Load-bearing premise

The whole argument rests on the claim that the filtered training corpus truly contains no multi-modifier noun phrases; if the dependency parser missed any, the models could have seen direct evidence of the target ordering and the poverty-of-stimulus framing would collapse.

Editorial extensions

If this is right

  • If the central claim is right, an innate language-specific constraint is not required for scope-homomorphic modifier order to emerge; a general sequence learner can retrieve the bias from indirect evidence.
  • The same filtered-corpus protocol can be applied to other proposed linguistic universals, turning each into a testable question about what can be learned when direct evidence is absent.
  • Because pointwise mutual information fails to explain model preferences, the source of the bias must lie in other distributional or structural properties of single-modifier input; identifying that property becomes the natural next step.
  • The cross-modifier contrast with human results (Dem-Num strongest in models, Dem-Adj strongest in humans) means any shared bias must be flexible in its surface strength, not a rigid universal ordering rule.
  • Model scale does not appear to be a prerequisite for this generalization: all three sizes show comparable preferences, so larger models are not needed to reproduce the effect.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial extension: a synthetic corpus with known noun-phrase structure would let researchers verify the filter's recall directly and test the same question without parser noise; this is the cleanest way to confirm the poverty-of-stimulus condition.
  • Editorial extension: if the training corpus's modifier classes were statistically permuted—for example, by making demonstratives the most noun-associated modifiers—the PMI account would predict a different preference profile; observing whether model preferences follow would distinguish structural from associative learning.
  • Editorial extension: applying the same procedure to a language whose dominant modifier order differs from English would test whether the learned preference tracks scope-homomorphism as a universal or merely English surface distributions.
  • Editorial extension: the near-chance PMI agreement leaves open the possibility that a context-sensitive association measure, rather than raw PMI, explains the item-level choices; this can be tested without changing the experimental design.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. The paper asks whether Transformer language models acquire the scope-homomorphic noun-phrase ordering bias under poverty-of-the-stimulus conditions. The authors filter a 100M-token Wikipedia sample by removing or perturbing sentences containing multi-modifier NPs, train OPT-style models of 52M, 110M, and 350M parameters with five seeds each, and evaluate them on minimal pairs of the form Dem-Num, Dem-Adj, and Num-Adj. They report HC_delta and HC% values above chance across all sizes, with Dem-Num strongest and Dem-Adj weakest, and then test whether noun-modifier PMI predicts item-level preferences, finding no reliable relationship. The central claim is that models develop a human-like ordering preference despite never observing multi-modifier NPs during training.

Significance. If the central premise is sound, this is a meaningful contribution: it extends Filtered Corpus Training to a well-known ALL bias, uses multi-seed and multi-size training with item-level statistics, and reports a clean null result for PMI. The paper also gives a controlled framework for asking which distributional cues are sufficient to induce scope-homomorphic preferences. However, the headline finding is conditional on the corpus filter's recall, which is not validated against any gold standard; the strength of the 'poverty of the stimulus' interpretation therefore depends entirely on a single, fallible parsing pipeline.

major comments (2)
  1. [§3.1, Appendix A] The Discussion's claim that the models 'never observ[e] noun phrases containing multiple modifiers during training' rests entirely on the Stanza-based filter. The manuscript reports that 148,367 of 6,032,739 sentences were flagged, but it never measures the filter's recall against a gold standard. The appendix itself documents a systematic parser failure ('that one' as SCONJ), patched by an explicit pattern filter, which demonstrates that the detection mechanism is imperfect; coordinated modifiers, appositive modifiers, and complex numeral/demonstrative strings are plausible additional miss classes. A false-negative rate of even 1% would leave more than a thousand multi-modifier NPs in the 100M-token corpus, and after 40 epochs the model would see each many times. Please provide a recall audit on a gold-standard sample of Wikipedia sentences, or explicitly restrict the claim to 'no detected multi-modifier NPs' and discuss the consequences of residual direct evidence for the main result. The Limitations section's concession that the corpus 'may contain indirect evidence about modifier relationships' should also be reconciled with the stronger formulation in the abstract and Discussion.
  2. [§3.4, §4.3] The evaluation corpus mixes 25 train-sourced and 25 test-sourced items per modifier pair, and the text says that test-sourced items may contain individual modifiers not seen during training. If these items contain lexical items that were not in the 100M-token training sample, they are not directly comparable to train-sourced items: the model could be making decisions about unseen word forms rather than about modifier ordering. This is especially relevant for the weak Dem-Adj and Num-Adj effects and for the mixed-effects model in Table 2. Please report train-sourced and test-sourced items separately, clarify how the test split was sampled relative to the training corpus, and state whether any modifier or noun tokens were out-of-vocabulary or unseen in training.
minor comments (3)
  1. [Appendix A] The text says 148,367 complex-NP sentences were 'filtered out' but then says they were 'perturbed' to create 664,272 rewritten variants that were included in the corpus; this wording is confusing and should be clarified so that readers can tell whether the original multi-modifier sentences were fully replaced by their single-modifier variants.
  2. [Figure 4] The caption contains a typo: 'Homomorphic consistency elta' should read 'Homomorphic consistency delta'.
  3. [§5.2.2] The item-level PMI analysis reports Pearson correlations and agreement percentages but no confidence intervals; given that the correlations are small and the agreement is near chance, reporting uncertainty around these estimates would make the null result more interpretable.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: LM preferences are direct empirical measurements; self-citations are background, and the parser-recall concern is a validity risk, not a circular step.

full rationale

The derivation chain is self-contained and empirical: the paper removes candidate multi-modifier NPs from training data (Section 3.1, Appendix A), trains Transformer LMs of three sizes, and directly measures log-probability differences between homomorphic and swapped modifier orders (Section 3.5, Eq. HC∆). No parameter is fitted to reproduce the target preference; the HC metric is defined independently of the training filter, and the PMI analysis (Section 5.2) tests an external hypothesis from Culbertson et al. (2020) and reports a null result, so it is not a post-hoc rationalization of the observed behavior. The self-citations (Patil et al. 2024; Shapiro and Steinert-Threlkeld 2023; Shapiro et al. 2024; Deng et al. 2026) provide background and method provenance for FiCT and iconic ALL, but the present result is not derived from those papers; the experiments stand on the authors' own training and evaluation pipeline. The paper's Limitations section honestly acknowledges that the training environment is not equivalent to human ALL and may contain indirect evidence, which further indicates that the authors are not claiming a definitional identity between input and output. The main remaining risk is the unvalidated recall of the Stanza-based filter: Appendix A admits the parser systematically misparses 'that one' and requires an explicit patch, so other missed multi-modifier constructions could in principle violate the 'never observing' premise. That is a correctness and external-validity concern, not circularity, because no equation or fitted quantity in the paper reduces the reported preference to the filter itself. Overall, the central claim is an empirical finding about model probabilities under a controlled manipulation, not a derivation that is equivalent to its inputs by construction.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the completeness of the corpus filter, the mapping of English canonical order to scope-homomorphism, the integrity of the train/test split, and the representativeness of the perturbed corpus for PMI estimation. None of these are derived in the paper; they are assumptions brought from the prior literature or the experimental setup.

assumptions (4)
  • domain assumption Stanza dependency parses fully identify all multi-modifier NPs of the three target types, so no direct ordering evidence survives in the training corpus.
    The poverty-of-stimulus claim depends on complete removal of Dem-Num, Dem-Adj, and Num-Adj multi-modifier NPs. The paper does not validate recall against a gold standard, and it explicitly notes one Stanza misparse pattern (Appendix A), raising the possibility of other missed constructions.
  • domain assumption English's canonical modifier order in the evaluation items is scope-homomorphic, so preferring the original over the swapped order is a preference for scope-homomorphism.
    The evaluation labels the corpus original as homomorphic. If some items in the 150 pairs do not follow the predicted hierarchy (e.g., idiomatic or lexicalized orders), the metric would misclassify preferences.
  • domain assumption The held-out 'test data' used to sample evaluation items was not included in the training corpus.
    The paper states items are sampled 25 from training and 25 from test data but never describes the train/test split; if the test items leaked into training, the 'unseen' evaluation is invalid.
  • domain assumption PMI estimates computed on the perturbed training corpus are representative of the original corpus's noun-modifier association strengths.
    The PMI analysis in Section 5.2 uses the artificially decomposed corpus, which differs from the original Wikipedia distribution. The paper does not verify that the perturbation preserves the relative PMI ordering (Adj > Num > Dem) that it relies on.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Language Models Generalize to Human-like Word Order Preferences." pith.science (2026). https://pith.science/paper/Z2A7SXW4

@misc{pith2026260805028,
  author       = {Pith},
  title        = {Pith review of: Language Models Generalize to Human-like Word Order Preferences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z2A7SXW4}},
  note         = {Machine review of arXiv:2608.05028}
}
read the original abstract

A central question in language acquisition is whether linguistic biases can emerge from general learning mechanisms operating over underdetermined input. Artificial Language Learning (ALL) studies have shown that human learners reliably generalize beyond the evidence provided, including by preferring scope-homomorphic noun phrase modifier orders. In this work, we investigate whether language models exhibit the same bias under similar conditions. We create a controlled learning environment in which models are trained on a corpus where all noun phrases containing multiple modifiers have been removed, eliminating direct evidence about modifier ordering, and are then evaluated on multiple modifier sentences. Across three model sizes, we find that they consistently prefer scope-homomorphic orders despite never observing them during training. These preferences vary in strength by modifier type. To investigate the source of these preferences, we examine noun-modifier association strength using pointwise mutual information (PMI). While PMI reflects known modifier-ordering patterns, it does not explain the models' ordering preferences. These findings demonstrate that LMs can recover human-like linguistic generalizations from impoverished input and provide a controlled framework for investigating the mechanisms underlying such biases.

Figures

Figures reproduced from arXiv: 2608.05028 by the authors.

Figure 1
Figure 1. A few possible examples of scope-homomorphic noun phrase orderings. The trees on the right depict the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Example of the corpus perturbation procedure for a Dem-Num noun phrase. Multi-modifier noun phrases [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Overall contextual delta and accuracy for homomorphic word orders across model sizes. All three models [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Homomorphic consistency elta and accuracy for homomorphic word orders by modifier type across model [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 42 canonical work pages

  1. [1]

    Substantive Learning Bias or an Effect of Familiarity?

    Goldberg, Adele , year = 2013, month = jun, journal =. Substantive Learning Bias or an Effect of Familiarity?. doi:10.1016/j.cognition.2013.02.017 , abstract =

    work page doi:10.1016/j.cognition.2013.02.017 2013

  2. [2]

    Structure

    Crain, Stephen and Nakayama, Mineharu , year = 1987, month = sep, journal =. Structure. doi:10.2307/415004 , urldate =

    doi:10.2307/415004 1987

  3. [3]

    Cognition , volume =

    The Learnability of Abstract Syntactic Principles , author =. Cognition , volume =. doi:10.1016/j.cognition.2010.11.001 , abstract =

    doi:10.1016/j.cognition.2010.11.001 2010

  4. [4]

    , year = 2005, journal =

    Reali, Florencia and Christiansen, Morten H. , year = 2005, journal =. Uncovering the. doi:10.1207/s15516709cog0000_28 , abstract =

    work page doi:10.1207/s15516709cog0000_28 2005

  5. [5]

    Cumulative Cultural Evolution in the Laboratory:

    Kirby, Simon and Cornish, Hannah and Smith, Kenny , year = 2008, journal =. Cumulative Cultural Evolution in the Laboratory:. doi:10.1073/pnas.0707835105 , abstract =

    doi:10.1073/pnas.0707835105 2008

  6. [6]

    Cognition , volume =

    Compression and Communication in the Cultural Evolution of Linguistic Structure , author =. Cognition , volume =. doi:10.1016/j.cognition.2015.03.016 , urldate =

    doi:10.1016/j.cognition.2015.03.016 2015

  7. [7]

    From Usage to Grammar:

    Bybee, Joan , year = 2006, journal =. From Usage to Grammar:. doi:10.1353/lan.2006.0186 , urldate =

    work page arXiv 2006

  8. [8]

    Behavioral and Brain Sciences , volume =

    Language Acquisition in the Absence of Experience , author =. Behavioral and Brain Sciences , volume =. doi:10.1017/S0140525X00071491 , abstract =

    work page doi:10.1017/s0140525x00071491

Show all 68 references
  1. [9]

    Proceedings of the National Academy of Sciences , volume =

    Language Learners Privilege Structured Meaning over Surface Frequency , author =. Proceedings of the National Academy of Sciences , volume =. doi:10.1073/pnas.1320525111 , urldate =

    doi

  2. [10]

    From the World to Word Order:

    Culbertson, Jennifer and Schouwstra, Marieke and Kirby, Simon , year = 2020, journal =. From the World to Word Order:

    2020

  3. [11]

    Artificial

    Culbertson, Jennifer , year = 2023, month = mar, publisher =. Artificial. doi:10.31234/osf.io/vg4mk , urldate =

    2023 doi

  4. [12]

    Frank, Robert and Mathis, Donald and Badecker, William , year = 2013, month = jul, journal =. The. doi:10.1080/10489223.2013.796950 , urldate =

    2013

  5. [13]

    , year = 2015, month = aug, publisher =

    O'Donnell, Timothy J. , year = 2015, month = aug, publisher =. Productivity and. doi:10.7551/mitpress/9780262028844.001.0001 , urldate =

    2015

  6. [14]

    Mission:

    Kallini, Julie and Papadimitriou, Isabel and Futrell, Richard and Mahowald, Kyle and Potts, Christopher , editor =. Mission:. Proceedings of the 62nd. doi:10.18653/v1/2024.acl-long.787 , urldate =

    2024 doi

  7. [15]

    Xu, Tianyang and Kuribayashi, Tatsuki and Oseki, Yohei and Cotterell, Ryan and Warstadt, Alex , year = 2025, month = feb, number =. Can. doi:10.48550/arXiv.2502.12317 , urldate =. arXiv , keywords =:2502.12317 , primaryclass =

    2025 doi arXiv

  8. [16]

    , booktitle =

    Qi, Peng and Zhang, Yuhao and Zhang, Yuhui and Bolton, Jason and Manning, Christopher D. , booktitle =. Stanza: A

  9. [17]

    30th Conference on Computational Natural Language Learning , year=

    Differences in Typological Alignment in Language Models' Treatment of Differential Argument Marking , author=. 30th Conference on Computational Natural Language Learning , year=

  10. [18]

    Testing learning hypotheses using neural networks by manipulating learning data , url =

    Leong, Cara Su-Yi and Linzen, Tal , month = jul, year =. Testing learning hypotheses using neural networks by manipulating learning data , url =. doi:10.48550/arXiv.2407.04593 , abstract =

    doi

  11. [19]

    , booktitle =

    Greenberg, Joseph H. , booktitle =. Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements , year =

  12. [20]

    Noisy-Context Surprisal as a Human Sentence Processing Cost Model , booktitle =

    Futrell, Richard and Levy, Roger , editor =. Noisy-Context Surprisal as a Human Sentence Processing Cost Model , booktitle =

  13. [21]

    and Brink, Kimberly and Bergen, Leon and Lim, Eunice and Saxe, Rebecca , year = 2013, month = jul, journal =

    Gibson, Edward and Piantadosi, Steven T. and Brink, Kimberly and Bergen, Leon and Lim, Eunice and Saxe, Rebecca , year = 2013, month = jul, journal =. A. doi:10.1177/0956797612463705 , urldate =

    2013 doi

  14. [22]

    Colorless

    Gulordava, Kristina and Bojanowski, Piotr and Grave, Edouard and Linzen, Tal and Baroni, Marco , editor =. Colorless. Proceedings of the 2018. doi:10.18653/v1/N18-1108 , urldate =

    2018 doi

  15. [23]

    Hahn, Michael and Degen, Judith and Goodman, Noah and Jurafsky, Dan and Futrell, Richard , abstract =. An

  16. [24]

    A Taxonomy and Review of Generalization Research in

    Hupkes, Dieuwke and Giulianelli, Mario and Dankers, Verna and Artetxe, Mikel and Elazar, Yanai and Pimentel, Tiago and Christodoulopoulos, Christos and Lasri, Karim and Saphra, Naomi and Sinclair, Arabella and Ulmer, Dennis and Schottmann, Florian and Batsuren, Khuyagbaatar an...

    2023

  17. [25]

    Language

    Jumelet, Jaap and Deni. Language. doi:10.48550/arXiv.2105.13818 , urldate =. arXiv , keywords =:2105.13818 , primaryclass =

    doi arXiv

  18. [26]

    Bigrams and the

    Kam, Xu. Bigrams and the. Cognitive Science , volume =. doi:10.1080/03640210802067053 , urldate =

    doi

  19. [27]

    Wilcox, Ethan Gotlieb and Futrell, Richard and Levy, Roger , year = 2024, month = oct, journal =. Using. doi:10.1162/ling_a_00491 , urldate =

    2024 doi

  20. [28]

    Jumelet, Jaap and Bylinina, Lisa and Zuidema, Willem and Szymanik, Jakub , year = 2026, month = feb, number =. Black. doi:10.48550/arXiv.2407.02136 , urldate =. arXiv , keywords =:2407.02136 , primaryclass =

    2026 doi

  21. [29]

    , year = 1980, publisher =

    Chomsky, N. , year = 1980, publisher =. Rules and

    1980

  22. [30]

    Aspects of the Theory of Syntax , author =

  23. [31]

    Martin, Alexander and Adger, David and Abels, Klaus and Kanampiu, Patrick and Culbertson, Jennifer , year = 2024, month = mar, journal =. A. doi:10.1177/09567976231222836 , urldate =

    2024 doi

  24. [32]

    Shapiro, Naomi Tachikawa and Hedding, Andrew and. Iconic. Proceedings of the Annual Meeting of the Cognitive Science Society , volume =

  25. [33]

    Decoupled

    Loshchilov, Ilya and Hutter, Frank , year = 2017, month = nov, journal =. Decoupled

    2017

  26. [34]

    doi:10.48550/arXiv.2205.01068 , urldate =

    Zhang, Susan and Roller, Stephen and Goyal, Naman and Artetxe, Mikel and Chen, Moya and Chen, Shuohui and Dewan, Christopher and Diab, Mona and Li, Xian and Lin, Xi Victoria and Mihaylov, Todor and Ott, Myle and Shleifer, Sam and Shuster, Kurt and Simig, Daniel and Koura, Puni...

    2022 doi arXiv

  27. [35]

    Thomas and Frank, Robert and Linzen, Tal , year = 2018, month = feb, journal =

    McCoy, R. Thomas and Frank, Robert and Linzen, Tal , year = 2018, month = feb, journal =. Revisiting the Poverty of the Stimulus: Hierarchical Generalization without a Hierarchical Bias in Recurrent Neural Networks , shorttitle =

    2018

  28. [36]

    Thomas and Frank, Robert and Linzen, Tal , editor =

    McCoy, R. Thomas and Frank, Robert and Linzen, Tal , editor =. Does. Transactions of the Association for Computational Linguistics , volume =. doi:10.1162/tacl_a_00304 , urldate =

    doi

  29. [37]

    Xu, Tianyang and Kuribayashi, Tatsuki and Oseki, Yohei and Cotterell, Ryan and Warstadt, Alex , year = 2026, journal =. Can. doi:10.1162/tacl.a.640 , urldate =

    2026 doi

  30. [38]

    Proceedings of the Annual Meeting of the Cognitive Science Society , volume =

    Developmental Evidence for Sensitivity to Hierarchical Structure in the Noun Phrase , author =. Proceedings of the Annual Meeting of the Cognitive Science Society , volume =

  31. [39]

    Assessing the

    Linzen, Tal and Dupoux, Emmanuel and Goldberg, Yoav , editor =. Assessing the. Transactions of the Association for Computational Linguistics , volume =. doi:10.1162/tacl_a_00115 , urldate =

    doi

  32. [40]

    doi:10.48550/arXiv.2007.06761 , urldate =

    Can Neural Networks Acquire a Structural Bias from Raw Linguistic Data? , author =. doi:10.48550/arXiv.2007.06761 , urldate =. arXiv , keywords =:2007.06761 , primaryclass =

    2007 doi arXiv

  33. [41]

    Learnability and the Statistical Structure of Language:

    Lewis,. Learnability and the Statistical Structure of Language:. Proceedings of the 26th Annual Boston University Conference on Language Development , number =

  34. [42]

    doi:10.31234/osf.io/jnqug , urldate =

    Cross-Linguistic Evidence for Cognitive Universals in the Noun Phrase , author =. doi:10.31234/osf.io/jnqug , urldate =

    doi

  35. [43]

    Glossa: a journal of general linguistics , volume =

    Experimental Evidence for the Influence of Structure and Meaning on Linear Order in the Noun Phrase , author =. Glossa: a journal of general linguistics , volume =. doi:10.5334/gjgl.1085 , urldate =

    doi

  36. [44]

    Targeted

    Marvin, Rebecca and Linzen, Tal , editor =. Targeted. Proceedings of the 2018. doi:10.18653/v1/D18-1151 , urldate =

    2018 doi

  37. [45]

    Thomas and Frank, Robert and Linzen, Tal , year = 2018, month = jun, number =

    McCoy, R. Thomas and Frank, Robert and Linzen, Tal , year = 2018, month = jun, number =. Revisiting the Poverty of the Stimulus: Hierarchical Generalization without a Hierarchical Bias in Recurrent Neural Networks , shorttitle =. doi:10.48550/arXiv.1802.09091 , urldate =. arXi...

    2018 doi arXiv

  38. [46]

    Constructing a

    Tomasello, Michael , year = 2003, eprint =. Constructing a. doi:10.2307/j.ctv26070v8 , urldate =

    2003 doi

  39. [47]

    , year = 1957, journal =

    Chomsky, N. , year = 1957, journal =. Syntactic

    1957

  40. [48]

    , author =

    A Review of Bf Skinner's Verbal Behavior. , author =. Language, 35 , pages =

  41. [49]

    Language

    Misra, Kanishka and Mahowald, Kyle , editor =. Language. Proceedings of the 2024. doi:10.18653/v1/2024.emnlp-main.53 , abstract =

    2024 doi

  42. [50]

    Filtered

    Patil, Abhinav and Jumelet, Jaap and Chiu, Yu Ying and Lapastora, Andy and Shen, Peter and Wang, Lexie and Willrich, Clevis and. Filtered. Transactions of the Association for Computational Linguistics , volume =. doi:10.1162/tacl_a_00720 , abstract =

    doi

  43. [51]

    Poverty of the

    Pearl, Lisa , year = 2022, month = oct, journal =. Poverty of the. doi:10.1080/15475441.2021.1981908 , urldate =

    2022

  44. [52]

    and Gibson, Edward , year = 2014, journal =

    Piantadosi, Steven T. and Gibson, Edward , year = 2014, journal =. Quantitative. doi:10.1111/cogs.12088 , urldate =

    2014 doi

  45. [53]

    Language Science Press , publisher =

    The Empirical Base of Linguistics , author =. Language Science Press , publisher =. doi:10.17169/langsci.b89.100 , urldate =

    doi

  46. [54]

    Hu, Jennifer and Gauthier, Jon and Qian, Peng and Wilcox, Ethan and Levy, Roger , editor =. A. Proceedings of the 58th. doi:10.18653/v1/2020.acl-main.158 , urldate =

    2020 doi

  47. [55]

    Subject-Verb Agreement with

    Wilson, Michael and Zhou, Zhenghao and Frank, Robert , editor =. Subject-Verb Agreement with. Proceedings of the

  48. [56]

    Wilcox, Ethan and Levy, Roger and Morita, Takashi and Futrell, Richard , editor =. What Do. Proceedings of the 2018. doi:10.18653/v1/W18-5423 , urldate =

    2018 doi

  49. [57]

    and Futrell, Richard and Mahowald, Kyle , editor =

    Papadimitriou, Isabel and Chi, Ethan A. and Futrell, Richard and Mahowald, Kyle , editor =. Deep. Proceedings of the 16th. doi:10.18653/v1/2021.eacl-main.215 , urldate =

    2021 doi

  50. [58]

    Adjective

    Scontras, Gregory , year = 2023, month = jan, journal =. Adjective. doi:10.1146/annurev-linguistics-030521-041835 , urldate =

    2023 doi

  51. [59]

    Assessing

    Goldberg, Yoav , year = 2019, month = jan, number =. Assessing. doi:10.48550/arXiv.1901.05287 , urldate =. arXiv , keywords =:1901.05287 , primaryclass =

    2019 doi arXiv

  52. [60]

    and Gebru, Timnit and

    Bender, Emily M. and Gebru, Timnit and. On the. Proceedings of the 2021. doi:10.1145/3442188.3445922 , urldate =

    2021

  53. [61]

    , year = 2018, journal =

    Dryer, Matthew S. , year = 2018, journal =. On the. 26630401 , eprinttype =

    2018

  54. [62]

    Deriving

    Cinque, Guglielmo , year = 2005, month = jul, journal =. Deriving. doi:10.1162/0024389054396917 , urldate =

    2005 doi

  55. [63]

    Shapiro, Naomi Tachikawa and. Iconic. Proceedings of the Annual Meeting of the Cognitive Science Society , volume =

  56. [64]

    , year = 2022, publisher =

    Warstadt, Alex and Bowman, Samuel R. , year = 2022, publisher =. What. Algebraic

    2022

  57. [65]

    Learning

    Wilson, Colin , year = 2006, journal =. Learning. doi:10.1207/s15516709cog0000_89 , urldate =

    2006 doi

  58. [66]

    Why Some Spatial Semantic Categories Are Harder to Learn than Others:

    Gentner, Dedre and Bowerman, Melissa , year = 2009, pages =. Why Some Spatial Semantic Categories Are Harder to Learn than Others:. Crosslinguistic Approaches to the Psychology of Language:

    2009

  59. [67]

    Manipulating Language Models' Training Data to Study Syntactic Constraint Learning:

    Leong, Cara Su-Yi and Linzen, Tal , year = 2026, month = aug, journal =. Manipulating Language Models' Training Data to Study Syntactic Constraint Learning:. doi:10.1016/j.jml.2026.104751 , urldate =

    2026

  60. [68]

    Anything

    Yang, Xiulin and Aoyama, Tatsuya and Yao, Yuekun and Wilcox, Ethan , year = 2025, month = may, number =. Anything. doi:10.48550/arXiv.2502.18795 , urldate =. arXiv , keywords =:2502.18795 , primaryclass =

    2025 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.